Improving Single-Engine Aircraft Safety Through Evolutionary Neural Networks

Abstract

Engine failure in single-engine aircraft is an uncommon yet critically dangerous occurrence that forces pilots to maximize glide distance before finding an optimal landing zone. While modern avionics sometimes provide emergency features which offer guidance to pilots, they often rely on static values calculated for the plane found in handbooks that fail to account for differences in air density and other external factors. Using a Python-based simulator, I investigate two approaches, a PID controller and an Evolutionary Neural Network based AI controller and compare them to human trials in maximizing glide distance. Results indicate that, compared to humans, both AI and PID controllers offer high-efficiency solutions to increase glide distance by up to 63.8% and, hence, survivability in such emergency aviation scenarios.